{"full_name":"jjang-ai/mlxstudio","owner":"jjang-ai","name":"mlxstudio","shard":"1","url":"https://github.com/jjang-ai/mlxstudio","description":"MLX Studio - Easiest way to run LLM's on your Mac. All in one engine.","homepage":"https://mlx.studio","topics":["llm","lmstudio","macbook","mlx","mlxllm","mlxstudio","ai","ai-agents","anthropic","anthropic-api","inference","inference-engine","openai-api","macstudio","apple-silicon"],"tag_pages":["ai","ai-agents","anthropic","apple-silicon","inference","llm","lmstudio","macbook","mlx","openai-api"],"language":null,"license":null,"avatar":"https://avatars.githubusercontent.com/u/138501656?u=fed2eea88b84d0a4561d0351fd1119cfd1037714&v=4","created_at":"2026-03-05T17:17:53Z","pushed_at":"2026-11-07T09:31:33Z","metrics":{"stars":991,"forks":64,"watchers":15,"issues":143,"prs":1,"releases":243,"issues_open":25,"issues_closed":118,"prs_open":0,"prs_closed":1,"prs_merged":0,"discussions":0},"growth":{"day":{"stars":0,"forks":0,"base_day":"2026-11-15"}},"history":{"dates":["2026-11-10","2026-11-11","2026-11-12","2026-11-13","2026-11-14","2026-11-15","2026-11-16"],"stars":[991,990,990,990,991,991,991],"forks":[64,64,64,64,64,64,64],"watchers":[null,15,15,15,15,15,15],"issues_open":[null,25,25,25,25,25,25],"prs_open":[null,0,0,0,0,0,0]},"boards":{},"same_owner":[],"issues":[{"number":117,"title":"[Bug] MiniMax-M2.7 JANGTQ_K: gibberish output regression introduced between v1.5.32 and v1.5.33","comments":6,"created_at":"2026-05-16T19:30:27Z"},{"number":75,"title":"Add ModelScope mirror support for users in mainland China","comments":5,"created_at":"2026-04-15T03:32:48Z"},{"number":31,"title":"Feature Request : Support for HTTP/SSE Transport and \"java\" command in STDIO for MCP settings","comments":5,"created_at":"2026-03-25T17:56:40Z"},{"number":121,"title":"Cannot load JANGQ-AI/Nemotron-3-Super-120B-A12B-JANG_2L","comments":3,"created_at":"2026-06-04T22:32:20Z"},{"number":132,"title":"LFM2.5-VL: converted bundles rename `feed_forward.w1/w2/w3` to `gate_proj/up_proj/down_proj` (are these models in scope?)","comments":1,"created_at":"2026-08-13T19:53:27Z"},{"number":67,"title":"Future Request if not already implemented: HotSwap of LoRa Adapters","comments":1,"created_at":"2026-04-13T19:35:24Z"},{"number":149,"title":"Feature request: System One decision-model endpoint (POST /v1/systemone)","comments":0,"created_at":"2026-11-07T14:33:50Z"},{"number":148,"title":"Royal star","comments":0,"created_at":"2026-11-01T06:41:50Z"}],"discussions":[],"latest_release":{"name":"vMLX 1.6.84","tag":"v1.6.84","published_at":"2026-11-07T09:29:01Z","body":"# vMLX 1.6.84\n\n## Changes\n\n- Publish vMLX 1.6.83 updater manifest (`b20e7513`)\n- feat(ricursive): Ripple-65B runtime and DFlash2 text lane (`f6bd17ff`)\n- fix(dflash2): text lane parity, rotating snapshot aliasing, drafter window (`0b33f432`)\n- fix(ripple): lane-owned max_tokens, history reasoning separators, launch test (`abdfeec3`)\n- perf(dflash2): lane matmul on text targets, exact Ripple verify attention (`aff7cfeb`)\n- perf(ripple): sliding-window SSD deltas; no thinking/effort controls (`9b6d00ad`)\n- test(dflash2): pin the lane matmul on text-only DFlash2 targets (`3e5a51d9`)\n- perf(dflash2): keep the finished turn hot for an immediate continuation (`2f765a65`)\n- perf(dflash2): cost-aware first observation cycle for codebook targets (`266f3bf4`)\n- perf(dflash2): turn entries chain one link deeper; the next boundary rebases (`5a34bcac`)\n- perf(dflash2): bound a windowed drafter's snapshot gap by its sliding window (`55630f99`)\n- perf(dflash2): amortized SSD chain rebase (`8a18ea30`)\n- perf(dflash2): a turn entry never pays the chain's whole rewrite (`645492a3`)\n- fix(tools): find_files matches path-style globs (`103d1d9a`)\n- perf(dflash2): Ripple gets its fastest settings from a plain load (`24e987e9`)\n- perf(dflash2): AR steps stop paying for a 16-row tile; probes run on a time budget (`9b24d4e3`)\n- perf(dflash2): split-K dense verify tile, drafter lane matmul, widths measured at load (`104007fa`)\n- panel: footer row per model request of a tool exchange (`f88fe5e0`)\n- panel: drop malformed fields from per-request footer rows (`9b94cf02`)\n- fix: report Ripple native DFlash2 cache in health and Perf (`83d0dca0`)\n- release: prepare Ripple scoped evidence gate for 1.6.84 (`89bbaa94`)\n\n## Runtime provenance\n\n- vMLX source: `8b3da66e11b87953758972a5cb4036ac1a982ddc`\n- JANG runtime: `44f84d1347a3692cc2d8ec61e8223535f7b99791` (`jang 2.5.53`)\n- Tahoe/macOS 26 is the default build; Sequoia is the compatibility build.\n- Both DMGs are Developer ID signed, independently notarized, stapled, and Gatekeeper checked by the candidate workflow.\n\n## Downloads\n\n- **Tahoe**: `vMLX-1.6.84-tahoe-arm64.dmg`\n  - SHA-256: `81c52ed49c15d69c3fd178d88dde2adeac094ea213069fd9daf174f13810edea`\n  - Size: `618360290` bytes\n- **Sequoia**: `vMLX-1.6.84-sequoia-arm64.dmg`\n  - SHA-256: `ae21c5c65661daf7c70fe331dd64f7f61d8c0bf10a5694a665f008a080df996f`\n  - Size: `582954755` bytes\n\n## Python distributions\n\n- Wheel: `vmlx-1.6.84-py3-none-any.whl` — SHA-256 `bc8534198e57cc820af0956366a82642a1424ca5dba2c16ef931f61b9b998a14`\n- Source: `vmlx-1.6.84.tar.gz` — SHA-256 `17b32344a578969520a9b005b3bd6c276569941b6816e7f95397463a6abac128`\n","is_prerelease":false},"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-16T09:23:14Z","history_start":"2026-09-06"}